BEBA research project: stateful processing for fast programmable switches
- BEBA project overview
- Related articles on this blog
- Main lines
- “I want to know more!”
BEBA project overview
At 6WIND we work on accelerating packets for network equipments. As for me, I work in the research and development team, and hence I realize little development on our core products. Instead, I am currently working on a European research project called BEBA. The name derives from BEhavioural BAsed forwarding, and indeed it aims at providing stateful packet processing capabilities to programmable switch.
To put it in another way, BEBA tries to bring back some of the “smartness” of the network architecture from the controllers to the switches. We have seen in a previous article how the controllers would program the switches, and how they would handle the “exception” packets: well, this model still applies with BEBA, but we try to make less of those packets considered as exceptions.
To that end, an abstraction layer has been devised so as to implement finite state machines on the switches. This text is a brief overview of the project, but I intend to write future articles about the technical contents, starting with details about this stateful packet processing.
Related articles on this blog
- Introduction to SDN — not directly related to BEBA.
- The present article, a general overview of the project.
- OpenState, the first version of the abstraction layer.
- BEBA’s use cases, a description of some of the principle use cases defined in the project.
- In-switch ARP with OpenState, an additional use case.
- Open Packet Processor, that extends OpenState with registers and boolean conditions on the latter.
- OpenState with eBPF, providing some details about my implementation of two use cases (port knocking and token bucket) for OpenState and Open Packet Processor, with eBPF.
- Hopefully, more to come!
Please note that all the articles of this series are some articles I wrote on my own, on my spare time, and that they are not endorsed by the BEBA consortium. They might contain inaccuracies or even mistakes, and in no case should their contents be considered as official statements from the BEBA consortium, nor from any of the project’s partners.
The work packages
BEBA project is split in a number of work packages, themselves divided in a number of tasks. Here is a summary list of the work packages.
WP1 is related to management and administrative stuff. Nothing technical here, so now let’s put it aside.
WP2 is a really interesting work package: it includes the conception, the definition and the validation, through proof-of-concept prototypes—be it software or hardware implementations—of the BEBA abstraction layer that is meant to provide stateful processing capabilities to the switches. So it is in this work package that the core mechanisms of the project are made explicit, and that the state machines are described in details. Some of those mechanisms should provide material for the next articles of this blog.
WP3 is based on WP2 and its prototypes, and brings them further: it consists in developing efficient and near-production implementations for the abstraction layer. Also, it encompasses another aspect of the project: besides bringing “smartness” to the switches, we want to bring them speed. So in this work package, we try to accelerate as much as possible—and a great part of what I do in the project takes place in here. So for this work package again, there should be a following article.
WP4 is about verification of the switches. It includes methods to assess and validate the correctness of the configuration, the security, the reliability and the resilience of the switches in the BEBA context.
WP5 includes an identification and a definition of several use cases for BEBA switches, that should be of interest for a great number of industrial entities. These use cases are considered either as “middlebox applications” (and they focusses on a particular processing that happens at the level of a switch, seen as a box) or as network-wide applications (and they rely on the interactions between several BEBA switches). I do not work specifically on this part. But the examples used for BEBA switches are fun to describe, so you may expect yet another article on this topic.
WP6 is for tests. We devise the switches, implement and accelerate them, find ways to verify their settings, and find uses for them, in the previous work packages. At last it will be time to check if all of this works! So there are two kinds of tests here, in short: in “laboratory”-like testbeds, and under real traffic conditions (thanks to CESNET, that may provide real traffic conditions for the tests).
WP7, at last, focuses on the dissemination of the concept that will have emerged from the project, in particular through scientific publications or presentations, standardization activities, or open-source contributions. The last important characteristic for the project is that the technologies introduced to improve the switches—in terms in functionalities as well as in speed—must be “vendor-agnostic”, meaning that they must not rely on the hardware or on a SDK provided by a single vendor. On the contrary, they should be easy to deploy on all platforms, hence the importance of standardization and open-source code.
The project started in January 2015, and runs until March 2017. At the time of this writing, the first half of the project has ended a couple of months ago.
At his point a question remains: Who’s in there? Here is the list of the partners involved in the project.
- CNIT (Consorzio Nazionale Interuniversitario per le Telecomunicazioni): an Italian research organism. They have many people working on SDN technologies, and are leaders of the BEBA project.
- KTH Royal Institute of Technology is a University in Stockholm, Sweden.
- CESNET is an operator for research and education in Czech Republic. They maintain a network for universities in the whole country; and they do research as well.
- NEC (German branch) is a big company selling IT products and services, including technologies related to SDN and packet processing.
- TCS (Thales Communication and Security) is a branch of the French Thales group, specialized in electronical, IT, defense, aerospace businesses (and probably a few others).
- And 6WIND, of course, is a medium French company, and your favorite partner for network acceleration solutions!
“I want to know more!”
All right, here are some additional leads. Of course, you can find all details on the project’s website. BEBA being a European project funded on the Horizon 2020 research plan, the results are public. In particular, you can consult the scientific articles published during the project, or even better if you want the very technical details of BEBA’s internals, the project deliveries that are all publicly accessible, save for the management part. It represents quite dense, but very exhaustive reading.
And also… why not read my other articles about BEBA?